On Fri, Apr 15, 2005 at 05:14:48PM -0400, Todd Vierling wrote:
> On Fri, 15 Apr 2005, Greg A. Woods wrote:
> 
> > If you take the time to examine how and where ${SETENV} has been and is
> > being used you'll find that each occurance sets _exactly_ and _only_ the
> > necessary values, which means that if one sanely uses '-i' then nothing
> > gets into the build of the package that's not explicitly supposed to be
> > there.
> 
> Am I finding myself agreeing with you again?  I really must stop that.  8-)
> 
> Actually, there's a really good point here.  If properly tested, this might
> be the best way to go.

Yeah, I know you're (both) in favour of this, and the last time it
came up, I remarked that I'd tried this before, and that it had not
worked correctly.  However, this was 6 years ago, on a Solaris and
NIS+ environment, and pkgsrc has changed in that time.  IIRC, the
failure was when the just-in-time su cut in, the su environment did
not retain enough information to be able to install files properly.

I know that Greg says that he's tried this and it's worked flawlessly,
but that doesn't mean to say it will work for everyone.  What I'd like
to hear from is people who have tried this on other environments.  If
we could do this properly with minimal intrusion, it would be a good
thing to have.

Oh, and please don't anyone change ${SETENV} to be ${ENV} - I'd have
thought the ksh ${ENV} would be reason enough to see why not, as has
already been mentioned.

Thanks,
Alistair